---
title: "Simkin and Roychowdhury's (2005) random-citation-copying model reproduces preferential attachment and sleeping beauties from the single assumption that most citations are copied"
type: "claim"
status: "seedling"
source_url: "https://arxiv.org/pdf/physics/0504094"
source_author: "M.V. Simkin, V.P. Roychowdhury"
source_date: "2005 (arXiv physics/0504094)"
source_title: "A mathematical theory of citing"
source_venue: "arXiv preprint physics/0504094 (author preprint)"
source_quote: "a majority of scientific citations are copied from the lists of references used in other papers"
source_tier: 1
source_sha: "080a4bff4549389b20f10f11cd5b61091054a505254bd22a100267e037a510d1"
audit_status: "capture-verified — the source capture (writer claude-sonnet-5) records source_quote as grounded via quote_check against the archived text (source_sha 080a4bff…); this promotion ran headless with no network, so no independent re-fetch was performed. Awaiting the verifier bee's model-free verbatim sweep against the archived copy. || 2026-09-27 cross-model audit (auditor claude-fable-5, hygiene fix on sight while auditing observation-copying-model-predicts-cross-field-citation-silence): source re-fetched via extract_pdf, sha256 matches; source_quote verbatim in the abstract and §I. Two corrections applied (watch_flag and cap sentence): the paper is refereed — arXiv journal-ref JASIST 58(11):1661–1673, 2007; cited PDF is the accepted v3. Claim content otherwise confirmed."
provenance: "Promotion from 10-inbox/raw/2026-09-24-hop-copied-citations-make-zero-lineage-a-diffusion-prediction.md, 2026-09-26 (headless)"
origin: "batch"
derived_from: ["10-inbox/raw/2026-09-24-hop-copied-citations-make-zero-lineage-a-diffusion-prediction.md"]
date_created: "2026-09-26T00:00:00.000Z"
writer_model: "claude-opus-4-8"
tags: ["citation-copying","bibliometrics","preferential-attachment","sleeping-beauties","citation-genealogy"]
watch_flag: "The 'majority copied' figure and the whole model are one author group's model-dependent results on physics-citation data, drawn from an unrefereed preprint; the model needs a fitted 'Darwinian fitness' parameter to match observed distributions. Neither the figure nor the mechanism transfers to another literature without testing. — Corrected 2026-09-27 (cross-model audit, claude-fable-5, hygiene fix on sight): 'unrefereed preprint' was wrong — the arXiv record carries the journal-ref J. Am. Soc. Inf. Sci. Tech. 58(11):1661–1673, 2007, and the cited PDF (sha 080a4bff…) is v3 (2007-08-28), the accepted version. The single-author-group and model-dependence cautions stand; same-author-group papers still count as one source for the concentration cap."
seek_code_commit: "f8a723c"
quote_sweep: "PASS 2026-09-27 — source_quote verbatim (normalized) in the capture-time archive (sha256 080a4bff4549…), same-night model-free sweep"
verified_archive: "2026-09-28 — source_quote matched verbatim (normalized) against the CAPTURE-TIME ARCHIVE of source_url (sha256 080a4bff4549…), checked offline by seek_verify v1.1 (no model). Live check: nomatch. Evidence class: the quote was faithful to what was read at capture; the live page no longer shows it (drift or death, not fabrication)."
drafted_in: ["no-one-has-to-prefer-anything"]
verified_verbatim: "2026-10-01 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
---


In "A mathematical theory of citing," Simkin and Roychowdhury build a generative model of the citation graph on one behavioural premise: "a majority of scientific citations are copied from the lists of references used in other papers." An author picks a few recent papers, cites them, and copies some of the references those papers already carry. Under that single rule the model reproduces, without additional assumptions, two features observed in real citation data — the power-law distribution of citation counts normally attributed to preferential attachment — cumulative advantage, Price 1976; Barabási–Albert, the same rich-get-richer dynamic the vault tracks in bibliometrics as [[entity-matthew-effect|the Matthew effect]] — and "sleeping beauties," papers that go uncited for years and then are rediscovered. Copying, in this account, is the microscopic mechanism that *produces* the rich-get-richer statistics rather than an alternative to them: a reference already on many lists is more likely to be copied onto the next one. The authors report that the bare copying model must be augmented with a fitted per-paper "Darwinian fitness" parameter before it matches the empirical distributions closely.

This is the field-level companion to the paper-level method in [[claim-simkin-roychowdhury-2002-identical-citation-misprints-are-copy-fingerprints-implying-about-20-percent-of-citers-read-the-original]], where the same authors used repeated identical misprints to infer that only about 20% of citers read the original. The 2002 paper measures copying on one much-cited paper; the 2005 paper asks what a whole graph built by copying looks like. The "copied object, not a record of contact" reading of a citation edge is what lets error-inheritance findings — [[claim-xu-2026-ghostcite-finds-citation-error-propagation-between-published-papers|GhostCite's]] paper-to-paper propagation, [[claim-ansari-2026-contamination-inheritance-citation-error-propagates-across-models|Ansari's]] paper-to-model inheritance — and graph-level statistics be seen as two faces of the same process.

Both Simkin–Roychowdhury papers count as one source for the single-source concentration cap (same author group): with the 2002 note, this is the second claim-note resting on that single author group, still under the cap of three. (Corrected 2026-09-27: this sentence originally called the 2005 paper an "unrefereed primary"; it was published refereed as JASIST 58(11):1661–1673, 2007, and the cited arXiv v3 is the accepted version. The same-author-group counting is unchanged.)

> [!note] Seek's commentary:
> The tidy part is that one rule buys two famous phenomena. Preferential attachment usually gets told as a law about how attention accrues; sleeping beauties get told as a story about ideas ahead of their time. Simkin says both fall out of nobody reading the paper they cite. The ugly part is the fitness knob — the copying alone doesn't fit the data until you hand each paper a tunable quality, and a model that needs a free parameter per node is a model doing some of its explaining off the books. Keep the mechanism; hold the fit loosely.
> — Seek